cjm-capability-pysbd

pySBD-based sentence-segmentation tool capability for the cjm-substrate runtime — rule-based Golden-Rules sentence boundaries delivered as character spans over the ORIGINAL text, trimmed to non-whitespace extents. Implements the sentence_segmentation task (cjm-sentence-segmentation-adapter-interface): pure compute, pure-pip, no model download, CPU-cheap — the segmenter behind the transcript decomposition pipeline's sentence-split stage (--sentence-split), chosen (DEC cc904eee) for the abbreviation/initial/honorific class ('Mr. Gorbachev', '3 p.m.', 'U.S.') that end-of-token heuristics mis-split. Config is a single language knob (Golden-Rules set selection, reload-triggered); clean=False + char_span=True are contract invariants, not knobs — spans must index into exactly the string the caller passed. Its effective-config hash joins the decomp skeleton-identity composite, so spines cut by different segmenter configs can never share node ids.

Modules

• cjm_capability_pysbd.__init__
• cjm_capability_pysbd.capability — Pure-compute sentence-segmentation tool capability using pySBD (Option C; B.5 work item 81e43606).

API

cjm_capability_pysbd.capability

• PySBDConfig class — Configuration for the pySBD segmenter.
• PySBDSegmentationCapability class — Sentence-segmentation tool capability using pySBD (pure compute).

Dependencies

Depends on: cjm-capability-primitives, cjm-substrate, pysbd